Research map / Time series / Long-term forecasting
TSFAdv: Frequency-Guided Black-Box Adversarial Attacks on Time Series Forecasting
Qizhuo Han, Xiangrui Cai, Sihan Xu, Ying Zhang, Zheli Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add profiled amplitude-phase sensitivity weights to frequency-domain NES (V3 vs V1) (ablation) | TSFAdv · 5 datasets average | DTW: 32.157 -> 43.940 (+37.0% DTW, +27.6% SME) | Improved |
| use random weights instead of profiled sensitivity weights (V2 vs V1) (ablation) | frequency-domain NES · 5 datasets average | DTW: 32.157 -> 16.708 | Hurt |
| move NES from time domain to frequency-domain parameterization alone (ablation) | NES · 5 datasets average | DTW: 36.32 -> 32.157 (11.5% reduction in DTW) | Hurt |
| use TSFAdv instead of NES for the attack | DLinear · Weather | DTW: 27.565 -> 77.161 (179.92%) | Improved |
| use TSFAdv instead of NES for the attack | TimesNet · Weather | DTW: 203.5007 -> 206.1333 (1.29%) | No clear effect |
| use TSFAdv instead of NES for the attack | DLinear · ETTm1 | DTW: 13.9518 -> 14.5333 (4.16%) | Mixed |
| double perturbation budget epsilon from 0.1 to 0.2 (ablation) | TSFAdv on TimeMixer · ETT, 200 queries | DTW: 150% DTW improvement (vs. 100% for NES) | Improved |
| apply randomized smoothing as defense | iTransformer under TSFAdv · ETT, 5 datasets | SME: 0.193 -> 0.266 | Hurt |
| apply magnitude pruning as defense | iTransformer under TSFAdv · 5 datasets average | DTW: 57.047 -> 58.118 | No clear effect |
| apply Gaussian-augmentation adversarial training as defense | PatchTST under TSFAdv · 5 datasets average | DTW: 38.355 -> 48.819 | Hurt |
| apply adversarial training as defense | iTransformer under TSFAdv · 5 datasets average | DTW: 57.047 -> 52.416 | Improved |
| use sensitivity profile from a different ETT dataset (h2->h1 vs h1->h1) (ablation) | TSFAdv · ETT domain | DTW: 34.169 vs. 34.190 | No clear effect |
Closest papers by meaning
- Battling the Non-stationarity in Time Series Forecasting via Test-time Adaptation (AAAI 2025 · similarity 0.85)
- Selective Learning for Deep Time Series Forecasting (NeurIPS 2025 · similarity 0.84)
- SynTSBench: Rethinking Temporal Pattern Learning in Deep Learning Models for Time Series (NeurIPS 2025 · similarity 0.84)
- Fortifying Time Series: DTW-Certified Robust Anomaly Detection (NeurIPS 2025 · similarity 0.84)
- BackTime: Backdoor Attacks on Multivariate Time Series Forecasting (NeurIPS 2024 · similarity 0.84)
- Multi-Scale Finetuning for Encoder-based Time Series Foundation Models (NeurIPS 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
